Deep learning has brought the most profound contribution towards biomedical image segmentation to automate the process of delineation in medical imaging. To accomplish such task, the models are required to be trained using huge amount of annotated or labelled data that highlights the region of interest with a binary mask. However, efficient generation of the annotations for such huge data requires expert biomedical analysts and extensive manual effort. It is a tedious and expensive task, while also being vulnerable to human error. To address this problem, a self-supervised learning framework, BT-Unet is proposed that uses the Barlow Twins approach to pre-train the encoder of a U-Net model via redundancy reduction in an unsupervised manner to learn data representation. Later, complete network is fine-tuned to perform actual segmentation. The BT-Unet framework can be trained with a limited number of annotated samples while having high number of unannotated samples, which is mostly the case in real-world problems. This framework is validated over multiple U-Net models over diverse datasets by generating scenarios of a limited number of labelled samples using standard evaluation metrics. With exhaustive experiment trials, it is observed that the BT-Unet framework enhances the performance of the U-Net models with significant margin under such circumstances.
This work presents a self-supervised learning framework named TeG to explore Temporal Granularity in learning video representations. In TeG, we sample a long clip from a video and a short clip that lies inside the long clip. We then extract their dense temporal embeddings. The training objective consists of two parts: a fine-grained temporal learning objective to maximize the similarity between corresponding temporal embeddings in the short clip and the long clip, and a persistent temporal learning objective to pull together global embeddings of the two clips. Our study reveals the impact of temporal granularity with three major findings. 1) Different video tasks may require features of different temporal granularities. 2) Intriguingly, some tasks that are widely considered to require temporal awareness can actually be well addressed by temporally persistent features. 3) The flexibility of TeG gives rise to state-of-the-art results on 8 video benchmarks, outperforming supervised pre-training in most cases.
Recent self-supervision methods have found success in learning feature representations that could rival ones from full supervision, and have been shown to be beneficial to the model in several ways: for example improving models robustness and out-of-distribution detection. In our paper, we conduct an empirical study to understand more precisely in what way can self-supervised learning - as a pre-training technique or part of adversarial training - affects model robustness to $l_2$ and $l_{\infty}$ adversarial perturbations and natural image corruptions. Self-supervision can indeed improve model robustness, however it turns out the devil is in the details. If one simply adds self-supervision loss in tandem with adversarial training, then one sees improvement in accuracy of the model when evaluated with adversarial perturbations smaller or comparable to the value of $\epsilon_{train}$ that the robust model is trained with. However, if one observes the accuracy for $\epsilon_{test} \ge \epsilon_{train}$, the model accuracy drops. In fact, the larger the weight of the supervision loss, the larger the drop in performance, i.e. harming the robustness of the model. We identify primary ways in which self-supervision can be added to adversarial training, and observe that using a self-supervised loss to optimize both network parameters and find adversarial examples leads to the strongest improvement in model robustness, as this can be viewed as a form of ensemble adversarial training. Although self-supervised pre-training yields benefits in improving adversarial training as compared to random weight initialization, we observe no benefit in model robustness or accuracy if self-supervision is incorporated into adversarial training.
A common vision from science fiction is that robots will one day inhabit our physical spaces, sense the world as we do, assist our physical labours, and communicate with us through natural language. Here we study how to design artificial agents that can interact naturally with humans using the simplification of a virtual environment. We show that imitation learning of human-human interactions in a simulated world, in conjunction with self-supervised learning, is sufficient to produce a multimodal interactive agent, which we call MIA, that successfully interacts with non-adversarial humans 75% of the time. We further identify architectural and algorithmic techniques that improve performance, such as hierarchical action selection. Altogether, our results demonstrate that imitation of multi-modal, real-time human behaviour may provide a straightforward and surprisingly effective means of imbuing agents with a rich behavioural prior from which agents might then be fine-tuned for specific purposes, thus laying a foundation for training capable agents for interactive robots or digital assistants. A video of MIA's behaviour may be found at https://youtu.be/ZFgRhviF7mY
Video self-supervised learning is a challenging task, which requires significant expressive power from the model to leverage rich spatial-temporal knowledge and generate effective supervisory signals from large amounts of unlabeled videos. However, existing methods fail to increase the temporal diversity of unlabeled videos and ignore elaborately modeling multi-scale temporal dependencies in an explicit way. To overcome these limitations, we take advantage of the multi-scale temporal dependencies within videos and proposes a novel video self-supervised learning framework named Temporal Contrastive Graph Learning (TCGL), which jointly models the inter-snippet and intra-snippet temporal dependencies for temporal representation learning with a hybrid graph contrastive learning strategy. Specifically, a Spatial-Temporal Knowledge Discovering (STKD) module is first introduced to extract motion-enhanced spatial-temporal representations from videos based on the frequency domain analysis of discrete cosine transform. To explicitly model multi-scale temporal dependencies of unlabeled videos, our TCGL integrates the prior knowledge about the frame and snippet orders into graph structures, i.e., the intra-/inter- snippet Temporal Contrastive Graphs (TCG). Then, specific contrastive learning modules are designed to maximize the agreement between nodes in different graph views. To generate supervisory signals for unlabeled videos, we introduce an Adaptive Snippet Order Prediction (ASOP) module which leverages the relational knowledge among video snippets to learn the global context representation and recalibrate the channel-wise features adaptively. Experimental results demonstrate the superiority of our TCGL over the state-of-the-art methods on large-scale action recognition and video retrieval benchmarks.The code is publicly available at https://github.com/YangLiu9208/TCGL.
Contrastive representation learning of videos highly relies on the availability of millions of unlabelled videos. This is practical for videos available on web but acquiring such large scale of videos for real-world applications is very expensive and laborious. Therefore, in this paper we focus on designing video augmentation for self-supervised learning, we first analyze the best strategy to mix videos to create a new augmented video sample. Then, the question remains, can we make use of the other modalities in videos for data mixing? To this end, we propose Cross-Modal Manifold Cutmix (CMMC) that inserts a video tesseract into another video tesseract in the feature space across two different modalities. We find that our video mixing strategy STC-mix, i.e. preliminary mixing of videos followed by CMMC across different modalities in a video, improves the quality of learned video representations. We conduct thorough experiments for two downstream tasks: action recognition and video retrieval on two small scale video datasets UCF101, and HMDB51. We also demonstrate the effectiveness of our STC-mix on NTU dataset where domain knowledge is limited. We show that the performance of our STC-mix on both the downstream tasks is on par with the other self-supervised approaches while requiring less training data.
Convolutional neural networks have turned into an illustrious tool for image fusion and super-resolution. However, their excellent performance cannot work without large fixed-paired datasets; and additionally, these high-demanded ground truth data always cannot be obtained easily in fusion tasks. In this study, we show that, the structures of generative networks capture a great deal of image feature priors, and then these priors are sufficient to reconstruct high-quality fused super-resolution result using only low-resolution inputs. By this way, we propose a novel self-supervised dataset-free method for adaptive infrared (IR) and visible (VIS) image super-resolution fusion named Deep Retinex Fusion (DRF). The key idea of DRF is first generating component priors which are disentangled from physical model using our designed generative networks ZipperNet, LightingNet and AdjustingNet, then combining these priors which captured by networks via adaptive fusion loss functions based on Retinex theory, and finally reconstructing the super-resolution fusion results. Furthermore, in order to verify the effectiveness of our reported DRF, both qualitative and quantitative experiments via comparing with other state-of-the-art methods are performed using different test sets. These results prove that, comparing with large datasets trained methods, DRF which works without any dataset achieves the best super-resolution fusion performance; and more importantly, DRF can adaptively balance IR and VIS information and has good noise immunity. DRF codes are open source available at https://github.com/GuYuanjie/Deep-Retinex-fusion.
As a basic task of computer vision, image similarity retrieval is facing the challenge of large-scale data and image copy attacks. This paper presents our 3rd place solution to the matching track of Image Similarity Challenge (ISC) 2021 organized by Facebook AI. We propose a multi-branch retrieval method of combining global descriptors and local descriptors to cover all attack cases. Specifically, we attempt many strategies to optimize global descriptors, including abundant data augmentations, self-supervised learning with a single Transformer model, overlay detection preprocessing. Moreover, we introduce the robust SIFT feature and GPU Faiss for local retrieval which makes up for the shortcomings of the global retrieval. Finally, KNN-matching algorithm is used to judge the match and merge scores. We show some ablation experiments of our method, which reveals the complementary advantages of global and local features.
The ability to generalize learned representations across significantly different visual domains, such as between real photos, clipart, paintings, and sketches, is a fundamental capacity of the human visual system. In this paper, different from most cross-domain works that utilize some (or full) source domain supervision, we approach a relatively new and very practical Unsupervised Domain Generalization (UDG) setup of having no training supervision in neither source nor target domains. Our approach is based on self-supervised learning of a Bridge Across Domains (BrAD) - an auxiliary bridge domain accompanied by a set of semantics preserving visual (image-to-image) mappings to BrAD from each of the training domains. The BrAD and mappings to it are learned jointly (end-to-end) with a contrastive self-supervised representation model that semantically aligns each of the domains to its BrAD-projection, and hence implicitly drives all the domains (seen or unseen) to semantically align to each other. In this work, we show how using an edge-regularized BrAD our approach achieves significant gains across multiple benchmarks and a range of tasks, including UDG, Few-shot UDA, and unsupervised generalization across multi-domain datasets (including generalization to unseen domains and classes).
This project focuses on the self-supervised training of convolutional neural networks (CNNs) and transformer networks for the task of image recognition. A simple siamese network with different backbones is used in order to maximize the similarity of two augmented transformed images from the same source image. In this way, the backbone is able to learn visual information without supervision. Finally, the method is evaluated on three image recognition datasets.
In this paper, we address a crucial problem in fashion e-commerce (with respect to customer experience, as well as revenue): color variants identification, i.e., identifying fashion products that match exactly in their design (or style), but only to differ in their color. We propose a generic framework, that leverages deep visual Representation Learning at its heart, to address this problem for our fashion e-commerce platform. Our framework could be trained with supervisory signals in the form of triplets, that are obtained manually. However, it is infeasible to obtain manual annotations for the entire huge collection of data usually present in fashion e-commerce platforms, such as ours, while capturing all the difficult corner cases. But, to our rescue, interestingly we observed that this crucial problem in fashion e-commerce could also be solved by simple color jitter based image augmentation, that recently became widely popular in the contrastive Self-Supervised Learning (SSL) literature, that seeks to learn visual representations without using manual labels. This naturally led to a question in our mind: Could we leverage SSL in our use-case, and still obtain comparable performance to our supervised framework? The answer is, Yes! because, color variant fashion objects are nothing but manifestations of a style, in different colors, and a model trained to be invariant to the color (with, or without supervision), should be able to recognize this! This is what the paper further demonstrates, both qualitatively, and quantitatively, while evaluating a couple of state-of-the-art SSL techniques, and also proposing a novel method.
Inspired by the recent success of self-supervised methods applied on images, self-supervised learning on graph structured data has seen rapid growth especially centered on augmentation-based contrastive methods. However, we argue that without carefully designed augmentation techniques, augmentations on graphs may behave arbitrarily in that the underlying semantics of graphs can drastically change. As a consequence, the performance of existing augmentation-based methods is highly dependent on the choice of augmentation scheme, i.e., hyperparameters associated with augmentations. In this paper, we propose a novel augmentation-free self-supervised learning framework for graphs, named AFGRL. Specifically, we generate an alternative view of a graph by discovering nodes that share the local structural information and the global semantics with the graph. Extensive experiments towards various node-level tasks, i.e., node classification, clustering, and similarity search on various real-world datasets demonstrate the superiority of AFGRL. The source code for AFGRL is available at https://github.com/Namkyeong/AFGRL.
Cold-start problem is a fundamental challenge for recommendation tasks. The recent self-supervised learning (SSL) on Graph Neural Networks (GNNs) model, PT-GNN, pre-trains the GNN model to reconstruct the cold-start embeddings and has shown great potential for cold-start recommendation. However, due to the over-smoothing problem, PT-GNN can only capture up to 3-order relation, which can not provide much useful auxiliary information to depict the target cold-start user or item. Besides, the embedding reconstruction task only considers the intra-correlations within the subgraph of users and items, while ignoring the inter-correlations across different subgraphs. To solve the above challenges, we propose a multi-strategy based pre-training method for cold-start recommendation (MPT), which extends PT-GNN from the perspective of model architecture and pretext tasks to improve the cold-start recommendation performance. Specifically, in terms of the model architecture, in addition to the short-range dependencies of users and items captured by the GNN encoder, we introduce a Transformer encoder to capture long-range dependencies. In terms of the pretext task, in addition to considering the intra-correlations of users and items by the embedding reconstruction task, we add embedding contrastive learning task to capture inter-correlations of users and items. We train the GNN and Transformer encoders on these pretext tasks under the meta-learning setting to simulate the real cold-start scenario, making the model easily and rapidly being adapted to new cold-start users and items. Experiments on three public recommendation datasets show the superiority of the proposed MPT model against the vanilla GNN models, the pre-training GNN model on user/item embedding inference and the recommendation task.
As an important branch in Recommender System, occasional group recommendation has received more and more attention. In this scenario, each occasional group (cold-start group) has no or few historical interacted items. As each occasional group has extremely sparse interactions with items, traditional group recommendation methods can not learn high-quality group representations. The recent proposed Graph Neural Networks (GNNs), which incorporate the high-order neighbors of the target occasional group, can alleviate the above problem in some extent. However, these GNNs still can not explicitly strengthen the embedding quality of the high-order neighbors with few interactions. Motivated by the Self-supervised Learning technique, which is able to find the correlations within the data itself, we propose a self-supervised graph learning framework, which takes the user/item/group embedding reconstruction as the pretext task to enhance the embeddings of the cold-start users/items/groups. In order to explicitly enhance the high-order cold-start neighbors' embedding quality, we further introduce an embedding enhancer, which leverages the self-attention mechanism to improve the embedding quality for them. Comprehensive experiments show the advantages of our proposed framework than the state-of-the-art methods.
We introduce a new setting of Novel Class Discovery in Semantic Segmentation (NCDSS), which aims at segmenting unlabeled images containing new classes given prior knowledge from a labeled set of disjoint classes. In contrast to existing approaches that look at novel class discovery in image classification, we focus on the more challenging semantic segmentation. In NCDSS, we need to distinguish the objects and background, and to handle the existence of multiple classes within an image, which increases the difficulty in using the unlabeled data. To tackle this new setting, we leverage the labeled base data and a saliency model to coarsely cluster novel classes for model training in our basic framework. Additionally, we propose the Entropy-based Uncertainty Modeling and Self-training (EUMS) framework to overcome noisy pseudo-labels, further improving the model performance on the novel classes. Our EUMS utilizes an entropy ranking technique and a dynamic reassignment to distill clean labels, thereby making full use of the noisy data via self-supervised learning. We build the NCDSS benchmark on the PASCAL-5$^i$ dataset and COCO-20$^i$ dataset. Extensive experiments demonstrate the feasibility of the basic framework (achieving an average mIoU of 49.81% on PASCAL-5$^i$) and the effectiveness of EUMS framework (outperforming the basic framework by 9.28% mIoU on PASCAL-5$^i$).
The label-embedded dictionary learning (DL) algorithms generate influential dictionaries by introducing discriminative information. However, there exists a limitation: All the label-embedded DL methods rely on the labels due that this way merely achieves ideal performances in supervised learning. While in semi-supervised and unsupervised learning, it is no longer sufficient to be effective. Inspired by the concept of self-supervised learning (e.g., setting the pretext task to generate a universal model for the downstream task), we propose a Self-Supervised Dictionary Learning (SSDL) framework to address this challenge. Specifically, we first design a $p$-Laplacian Attention Hypergraph Learning (pAHL) block as the pretext task to generate pseudo soft labels for DL. Then, we adopt the pseudo labels to train a dictionary from a primary label-embedded DL method. We evaluate our SSDL on two human activity recognition datasets. The comparison results with other state-of-the-art methods have demonstrated the efficiency of SSDL.
This paper proposes a probabilistic contrastive loss function for self-supervised learning. The well-known contrastive loss is deterministic and involves a temperature hyperparameter that scales the inner product between two normed feature embeddings. By reinterpreting the temperature hyperparameter as a quantity related to the radius of the hypersphere, we derive a new loss function that involves a confidence measure which quantifies uncertainty in a mathematically grounding manner. Some intriguing properties of the proposed loss function are empirically demonstrated, which agree with human-like predictions. We believe the present work brings up a new prospective to the area of contrastive learning.
Self-supervised learning aims to learn image feature representations without the usage of manually annotated labels. It is often used as a precursor step to obtain useful initial network weights which contribute to faster convergence and superior performance of downstream tasks. While self-supervision allows one to reduce the domain gap between supervised and unsupervised learning without the usage of labels, the self-supervised objective still requires a strong inductive bias to downstream tasks for effective transfer learning. In this work, we present our material and texture based self-supervision method named MATTER (MATerial and TExture Representation Learning), which is inspired by classical material and texture methods. Material and texture can effectively describe any surface, including its tactile properties, color, and specularity. By extension, effective representation of material and texture can describe other semantic classes strongly associated with said material and texture. MATTER leverages multi-temporal, spatially aligned remote sensing imagery over unchanged regions to learn invariance to illumination and viewing angle as a mechanism to achieve consistency of material and texture representation. We show that our self-supervision pre-training method allows for up to 24.22% and 6.33% performance increase in unsupervised and fine-tuned setups, and up to 76% faster convergence on change detection, land cover classification, and semantic segmentation tasks.
Graph Neural Networks (GNNs) have achieved great success in processing graph data by extracting and propagating structure-aware features. Existing GNN research designs various propagation schemes to guide the aggregation of neighbor information. Recently the field has advanced from local propagation schemes that focus on local neighbors towards extended propagation schemes that can directly deal with extended neighbors consisting of both local and high-order neighbors. Despite the impressive performance, existing approaches are still insufficient to build an efficient and learnable extended propagation scheme that can adaptively adjust the influence of local and high-order neighbors. This paper proposes an efficient yet effective end-to-end framework, namely Contrastive Adaptive Propagation Graph Neural Networks (CAPGNN), to address these issues by combining Personalized PageRank and attention techniques. CAPGNN models the learnable extended propagation scheme with a polynomial of a sparse local affinity matrix, where the polynomial relies on Personalized PageRank to provide superior initial coefficients. In order to adaptively adjust the influence of both local and high-order neighbors, a coefficient-attention model is introduced to learn to adjust the coefficients of the polynomial. In addition, we leverage self-supervised learning techniques and design a negative-free entropy-aware contrastive loss to explicitly take advantage of unlabeled data for training. We implement CAPGNN as two different versions named CAPGCN and CAPGAT, which use static and dynamic sparse local affinity matrices, respectively. Experiments on graph benchmark datasets suggest that CAPGNN can consistently outperform or match state-of-the-art baselines. The source code is publicly available at https://github.com/hujunxianligong/CAPGNN.
Advances in generative modeling based on GANs has motivated the community to find their use beyond image generation and editing tasks. In particular, several recent works have shown that GAN representations can be re-purposed for discriminative tasks such as part segmentation, especially when training data is limited. But how do these improvements stack-up against recent advances in self-supervised learning? Motivated by this we present an alternative approach based on contrastive learning and compare their performance on standard few-shot part segmentation benchmarks. Our experiments reveal that not only do the GAN-based approach offer no significant performance advantage, their multi-step training is complex, nearly an order-of-magnitude slower, and can introduce additional bias. These experiments suggest that the inductive biases of generative models, such as their ability to disentangle shape and texture, are well captured by standard feed-forward networks trained using contrastive learning. These experiments suggest that the inductive biases present in current generative models, such as their ability to disentangle shape and texture, are well captured by standard feed-forward networks trained using contrastive learning.